Constrained Partitioning of Sparse Matrices Using Graph Decompositions

PONZONI, IGNACIO; SÁNCHEZ, MABEL C.; BRIGNOLE, NÉLIDA B.

Toronto, Canadá

Conferencia; SIAM Annual Meeting 1998; 1998

SIAM (Society for Industrial and Applied Mathematics)

A direct method based on graph decompositions that car­ries out structural rearrangements of huge sparse occur­rence matrices, also comprising structurally singular ma­trices, was developed. The procedure yields a Block lower-Triangular Form specially adapted to solve plant in­strumentation problems. The decomposition is obtained through a new vertex classification. Another innovative feature is the incorporation of partitioning constraints to ensure realistic results. The final decomposition is opti­mum because it yields the maximum number of irreducible blocks.